first E2E working implementation

test.json

@@ -1,61 +0,0 @@
-[
-  {
-    "$schema": "https://json-schema.org/draft/2020-12/schema",
-    "$id": "https://www.enricorampazzo.tech/dam_helper.schema.json",
-    "title": "Answer",
-    "description": "an answer to a question",
-    "type": "array",
-    "items": {
-      "type": "object",
-      "properties": {
-        "question": {
-          "type": "string"
-        },
-        "answer": {
-          "type": ["string", "null"]
-        }
-      },
-      "required": ["question", "answer"]
-    }
-  },
-  {
-    "question": "1) What do you need to do?",
-    "answer": "Pest control"
-  },
-  {
-    "question": "2) In which community is the work taking place?",
-    "answer": "JBR"
-  },
-  {
-    "question": "3) In which building?",
-    "answer": "Sadaf 5"
-  },
-  {
-    "question": "4) In which unit/apartment number?",
-    "answer": "2408"
-  },
-  {
-    "question": "5) Am I an owner or a tenant?",
-    "answer": "unclear"
-  },
-  {
-    "question": "6) In which date is the work taking place?",
-    "answer": "7/9/2024"
-  },
-  {
-    "question": "7) In which date will the work finish?",
-    "answer": null
-  },
-  {
-    "question": "8) What is my contact number?",
-    "answer": null
-  },
-  {
-    "question": "9) What is the name of the contracting company?",
-    "answer": "Breathe Maintenance"
-  },
-  {
-    "question": "10) What is the contact number of the contracting company?",
-    "answer": null
-  }
-]

.gitignore

@@ -1,4 +1,3 @@
 .idea
-form.pdf
 /.idea/
 /.idea/inspectionProfiles/

app.py

@@ -1,78 +1,45 @@
-import json
-import re
-from json import JSONDecodeError
+from pathlib import Path
 
+from prompts.prompts_manager import PromptsManager
 from repository.ollama import OllamaRepository
-from schema import ModelRoles
+from llm.llm import ModelRoles, Model
+from form.form import work_categories, build_form_data_from_answers, write_pdf_form
 
-# this regex is most definitely *not* going to make my system blow up unexpectedly at some point in the future
-get_questions_and_answers_regex = re.compile(r'{\s*"question":\s*"([^"]+)",\s*"answer":\s*(\s*null\s*|"[^"]+"\s*)}')
 
+def check_for_missing_answers(parsed_questions: dict[int, str]):
+    return [k for k in parsed_questions if parsed_questions[k] is None]
 
-def print_current_status(total_success, first_try_successes, second_try_successes, regex_successes):
-    print(f"""so far, questions and answers were parsed successfully {total_success} times, of which 
-              {first_try_successes} were successful at the first attempt, 
-              {second_try_successes} were successful after asking the model to self-correct with a 'JSON expert' agent 
-              and {regex_successes} were successful using a regex
-          """)
+
+def ask_again(missing_questions: list[int], user_questions: list[str], parsed_questions: dict[int, str]):
+    for id_ in missing_questions:
+        answer = input(f"I could not find the answer to this question: {user_questions[id_].lower()}")
+        parsed_questions[id_] = answer
 
 
 if __name__ == '__main__':
-    with open("questions.txt") as questions_file:
-        questions = questions_file.read()
-    with open("system_prompt.txt") as system_prompt_file:
-        system_prompt = system_prompt_file.read()
-    with open("verification_prompt.txt") as verification_prompt_file:
-        verification_prompt = verification_prompt_file.read()
-    verification_prompt = verification_prompt.replace("{questions}", questions)
+    prompts_manager = PromptsManager()
     user_prompt = input(f"Please describe what you need to do. To get the best results "
-          f"try to answer the following questions:\n{questions}\n\n>")
-
-    ollama_repository = OllamaRepository("llama3.1", system_prompt,
-                                         ModelRoles("system", "user", "assistant"))
-    successful = 0
-    corrected_json_was_ok = 0
-    regex_got_all_answers = 0
-    successful_at_first_attempt = 0
-
-    while True:
-        ollama_repository.send_prompt(f"Ingest the following information: {user_prompt}")
-        answer = ollama_repository.send_prompt(verification_prompt)
-        ollama_repository.system_msg = "You are an expert at JSON format, and can detect and fix errors in JSON strings"
-        fixed_json = ollama_repository.send_prompt(f"Verify if this is legal JSON and correct any mistake, output just the fixed json without adding anything else {answer['content']}")
-        answers = fixed_json["content"]
-        try:
-            json_answers = json.loads(answers, strict=False)
-            print(answers)
-            successful_at_first_attempt +=1
-            successful+=1
-            print_current_status(successful, successful_at_first_attempt, corrected_json_was_ok, regex_got_all_answers)
-        except JSONDecodeError as e:
-            print("there was a problem in this json, asking the LLM to fix it passing the exception error message")
-            answer = (
-                ollama_repository.send_prompt(f"When parsing this json {answers} I got this error {str(e)}. "
-                                              f"Fix this error and return just the corrected json without adding anything else"))
-
-            print("trying to parse the corrected json")
-            try:
-                json_answers = json.loads(answer["content"])
-                corrected_json_was_ok+=1
-                print_current_status(successful, successful_at_first_attempt, corrected_json_was_ok,
-                                     regex_got_all_answers)
-            except JSONDecodeError as e:
-                print("still error, going old school")
-                regex_parse_result = {}
-                for match in get_questions_and_answers_regex.findall(answer["content"]):
-                    question, answer = match
-                    regex_parse_result[question] = answer
-                if len(regex_parse_result) == 10:
-                    print("I got all 10 answers apparently")
-                    successful+=1
-                    regex_got_all_answers+=1
-                    print_current_status(successful, successful_at_first_attempt, corrected_json_was_ok,
-                                         regex_got_all_answers)
-                else:
-                    print(f"unable to parse \n {answers}\n giving up")
-                    break
-
-
+                        f"try to answer all the following questions:\n{'\n'.join(prompts_manager.questions)}\n\n>")
+
+    ollama_repository = OllamaRepository(Model("llama3.1",
+                                               ModelRoles("system", "user", "assistant")),
+                                         prompts_manager.system_prompt,
+                                         )
+    ollama_repository.send_prompt(f"Ingest the following information: {user_prompt}")
+    answers = {}
+    for idx, q in enumerate(prompts_manager.verification_prompt):
+        answer = ollama_repository.send_prompt(
+            f"Answer the following question, if the answer is not present just answer null. Keep the answer brief: {q}")
+        answers[idx] = None if 'null' in answer["content"].lower() else answer['content']
+    missing_answers = check_for_missing_answers(answers)
+    while missing_answers:
+        ask_again(missing_answers, prompts_manager.questions, answers)
+        missing_answers = check_for_missing_answers(answers)
+    answer = ollama_repository.send_prompt(
+        f"The work to do is {answers[1]}. Given the following categories {work_categories.values()} which ones are the most relevant? Only return one categories, separated by a semicolon")
+    categories = []
+    for category in answer["content"].split(";"):
+        categories.extend([k for k, v in work_categories.items() if category in v])
+
+    form_data = build_form_data_from_answers(answers, categories, f"{Path(__file__, "..", "signature.png")}")
+    write_pdf_form(form_data, Path("signed_form1.pdf"))

schema.py → form/form.py

@@ -1,5 +1,40 @@
+from pathlib import Path
 from typing import TypedDict
+from PyPDFForm import PdfWrapper
 
+from utils.date_utils import get_today_date_as_dd_mm_yyyy
+from utils.parsing_utils import find_and_parse_date, find_and_parse_phone_number
+
+
+def build_form_data_from_answers(answers: dict[int, str], categories: list[str], signature: str | None = None):
+    form_data = {}
+    for category in categories:
+        form_data[form_fields_map[category]] = True
+    form_data[form_fields_map["minor_work"]] = True
+    form_data[form_fields_map["signed_by"]] = answers[0]
+    form_data[form_fields_map["community"]] = answers[2]
+    form_data[form_fields_map["building_name"]] = answers[3]
+    form_data[form_fields_map["unit_no"]] = answers[4]
+    form_data[form_fields_map["owner" if "owner" in answers[5].lower() else "tenant"]] = True
+    form_data[form_fields_map["start_date"]] = find_and_parse_date(answers[6])
+    form_data[form_fields_map["end_date"]] = find_and_parse_date(answers[7])
+    form_data[form_fields_map["contact_number"]] = find_and_parse_phone_number(answers[8])
+    form_data[form_fields_map["contractor_name"]] = answers[9]
+    form_data[form_fields_map["contractor_contact_number"]] = answers[10]
+    form_data[form_fields_map["contractor_email"]] = answers[11]
+    form_data[form_fields_map["email"]] = answers[12]
+    form_data[form_fields_map["occupant_name"]] = answers[0]
+    form_data[form_fields_map["signature_date"]] = get_today_date_as_dd_mm_yyyy()
+    if signature:
+        form_data[form_fields_map["signature"]] = signature
+    return form_data
+
+def write_pdf_form(form_data: dict[str, str | bool | int], output_path:Path):
+    base_form_path = Path(__file__, "..", "form.pdf")
+    with open(base_form_path, "rb") as blank_form:
+        filled_form = PdfWrapper(blank_form).fill(form_data)
+    with open(output_path, "wb+") as filled_form_file:
+        filled_form_file.write(filled_form.read())
 
 class FormFields(TypedDict, total=True):
     start_date: str
@@ -44,55 +79,6 @@ class FormFields(TypedDict, total=True):
     minor_work: str
 
 
-class MinorWorkForm:
-    def __init__(self):
-        self.start_date: str
-        self.end_date: str
-        self.signed_by: str
-        self.signature_date: str
-        self.community: str
-        self.email: str
-        self.unit_no: str
-        self.contact_number: str
-        self.building_name: str
-        self.contractor_name: str
-        self.contractor_email: str
-        self.contractor_contact_number: str
-        self.occupant_name: str
-        self.civil: bool
-        self.electrical_AC: bool
-        self.furniture_delivery: bool
-        self.handyman: bool
-        self.pest_control: bool
-        self.sail_shade: bool
-        self.wardrobes: bool
-        self.wooden_flooring: bool
-        self.plumbing_sanitary: bool
-        self.safety_systems: bool
-        self.soft_landscaping: bool
-        self.balcony_tiles: bool
-        self.bathroom_toilet_refurbish: bool
-        self.staircase_railing: bool
-        self.storage_box: bool
-        self.tiling_works: bool
-        self.doors_door_frames: bool
-        self.gypsum_ceiling: bool
-        self.kitchen_refurbish: bool
-        self.lights_and_sockets: bool
-        self.painting_wallpapering: bool
-        self.seating_area_barbecuing_fountain: bool
-        self.other_work_checkbox: bool
-        self.owner: bool
-        self.tenant: bool
-        self.signature: str
-
-class ModelRoles:
-    def __init__(self, system_role, user_role, ai_role):
-        self.system_role = system_role
-        self.user_role = user_role
-        self.ai_role = ai_role
-
-
 form_fields_map: FormFields = {"start_date": "From", "end_date": "to", "signed_by": "name-10",
                                "signature_date": "From--1", "community": "Name-2", "email": "Name-3",
                                "unit_no": "Name-4", "contact_number": "Name-5", "building_name": "Name-6",
@@ -101,11 +87,28 @@ form_fields_map: FormFields = {"start_date": "From", "end_date": "to", "signed_b
                                "other_work": "name-19", "civil": "3", "electrical_AC": "10", "furniture_delivery": "14",
                                "handyman": "18", "pest_control": "21", "sail_shade": "12", "wardrobes": "16",
                                "wooden_flooring": "24",
-                               "plumbing_sanitary": "4", "safety systems": "4", "soft_landscaping": "9",
+                               "plumbing_sanitary": "4", "safety_systems": "4", "soft_landscaping": "9",
                                "balcony_tiles": "13", "bathroom_toilet_refurbish": "20",
                                "staircase_railing": "17", "storage_box": "11", "tiling_works": "15",
                                "doors_door_frames": "6", "gypsum_ceiling": "22", "kitchen_refurbish": "7",
                                "lights_and_sockets": "23", "painting_wallpapering": "19",
                                "seating_area_barbecuing_fountain": "8",
                                "other_work_checkbox": "69", "owner": "Check Box4", "tenant": "Check Box3",
-                               "signature": "Signature2", "minor_work": "1"}
+                               "signature": "Signature2", "minor_work": "2"}
+work_categories = {"other_work": "other work", "civil": "Civil work, masonry",
+                   "electrical_AC": "electrical / AC / air conditioning", "furniture_delivery": "furniture delivery",
+                   "handyman": "handyman work, repairs, furniture assembly, furniture installation", "pest_control": "pest control",
+                   "sail_shade": "sail shade, curtains", "wardrobes": "wardrobes",
+                   "wooden_flooring": "wooden flooring",
+                   "plumbing_sanitary": "plumbing / sanitary / bathroom works such as repairing a sink, tap, toilet, shower or bathtub",
+                   "safety_systems": "safety systems, surveillance systems installation",
+                   "soft_landscaping": "soft landscaping", "balcony_tiles": "installation and fixing of balcony tiles",
+                   "bathroom_toilet_refurbish": "bathroom or toilet refurbishment or renovation",
+                   "staircase_railing": "staircase rails, handrails", "storage_box": "storage box",
+                   "tiling_works": "tiles installation or repair",
+                   "doors_door_frames": "work related to doors and door frames",
+                   "gypsum_ceiling": "gypsum ceiling installation or maintenance",
+                   "kitchen_refurbish": "kitchen renovation or refurbishment",
+                   "lights_and_sockets": "installation of lights, ceiling lights, power sockets",
+                   "painting_wallpapering": "painting, wallpaper installation",
+                   "seating_area_barbecuing_fountain": "installation or renovation of outdoor structures such as seating area, barbecuing area or fountains"}

llm/llm.py

@@ -0,0 +1,11 @@
+class ModelRoles:
+    def __init__(self, system_role: str, user_role: str, ai_role: str):
+        self.system_role: str = system_role
+        self.user_role: str = user_role
+        self.ai_role: str = ai_role
+
+
+class Model:
+    def __init__(self, name: str, roles: ModelRoles):
+        self.name: str = name
+        self.roles: ModelRoles = roles

prompts/prompts_manager.py

@@ -0,0 +1,19 @@
+import datetime
+from pathlib import Path
+
+from utils.date_utils import get_today_date_as_dd_mm_yyyy
+
+
+class PromptsManager:
+    def __init__(self):
+        base_path = Path(__file__, "..")
+        with open(Path(base_path, "system_prompt.txt")) as sysprompt_file:
+            self.system_prompt: str = sysprompt_file.read()
+        with open(Path(base_path, "questions.txt")) as questions_file:
+            self.questions: list[str] = questions_file.readlines()
+        with open(Path(base_path, "verification_prompt2.txt")) as verification_prompt_file:
+            verification_prompt = verification_prompt_file.readlines()
+            todays_date = get_today_date_as_dd_mm_yyyy()
+            for line in verification_prompt:
+                line.replace("{today}", todays_date)
+            self.verification_prompt: list[str] = verification_prompt

prompts/questions.txt

@@ -0,0 +1,13 @@
+What is your full name?
+What is the nature of the work you need to do?
+In which community is the work taking place?
+In which building?
+In which unit/apartment number?
+Are you an owner or a tenant?
+In which date is the work taking place?
+In which date will the work finish?
+What is your contact number?
+What is the name of the contracting company?
+What is the contact number of the contracting company?
+What is the email of the contracting company?
+What is your email?

prompts/verification_prompt2.txt

@@ -0,0 +1,13 @@
+What is my full name?
+What is the nature of the work I need to do?
+In which community is the work taking place?
+In which building?
+In which unit/apartment number?
+Am I the owner or the tenant?
+In which date is the work taking place? Please answer with just a date formatted as dd/mm/yyyy. In case I used expressions like today, tomorrow, in two days, ecc, know that today it is 08/09/2024
+In which date will the work finish? Please answer with just a date formatted as dd/mm/yyyy. In case I used expressions like today, tomorrow, in two days, ecc, know that today it is {today}. If no date is provided, consider that it will finish on the same day as the start date
+What is my contact number?
+What is the name of the contracting company?
+What is the contact number of the contracting company?
+What is the email of the contracting company?
+What is your email?

questions.txt

@@ -1,10 +0,0 @@
-1) What do you need to do?
-2) In which community is the work taking place?
-3) In which building?
-4) In which unit/apartment number?
-5) Am I an owner or a tenant?
-6) In which date is the work taking place?
-7) In which date will the work finish?
-8) What is my contact number?
-9) What is the name of the contracting company?
-10) What is the contact number of the contracting company?

repository/ollama.py

@@ -1,21 +1,22 @@
 import ollama
 from ollama import Options
 
-from schema import ModelRoles
+from llm.llm import ModelRoles, Model
 
 
 class OllamaRepository:
-    def __init__(self, model, system_msg, roles: ModelRoles):
-        self.model = model
-        self.system_msg = system_msg
-        self.roles = roles
-        self.message_history: list[dict[str, str]] = [{"role": self.roles.system_role, "content": system_msg}]
+    def __init__(self, model:Model, system_msg):
+        self.model: Model = model
+        self.system_msg: str = system_msg
+        self.message_history: list[dict[str, str]] = [{"role": self.model.roles.system_role, "content": system_msg}]
 
-    def send_prompt(self, prompt:str, add_to_history:bool = False) -> dict[str, str]:
-        options: Options = Options(temperature=0.1)
-        self.message_history.append({"role": self.roles.user_role, "content":prompt})
-        response = ollama.chat(self.model, self.message_history, options=options)
-        answer = {"role": self.roles.ai_role, "content": response["message"]["content"]}
+    def send_prompt(self, prompt:str, add_to_history:bool = True) -> dict[str, str]:
+        options: Options = Options(temperature=0)
+        self.message_history.append({"role": self.model.roles.user_role, "content":prompt})
+        response = ollama.chat(self.model.name, self.message_history, options=options)
+        answer = {"role": self.model.roles.ai_role, "content": response["message"]["content"]}
         if add_to_history:
             self.message_history.append(answer)
+        else:
+            self.message_history.pop()
         return answer

utils/date_utils.py

@@ -0,0 +1,3 @@
+import datetime
+def get_today_date_as_dd_mm_yyyy():
+    return datetime.datetime.now().strftime("%d/%m/%Y")

utils/parsing_utils.py

@@ -0,0 +1,17 @@
+import re
+parse_date_regex = re.compile(r"([0-9]{1,2}/[0-9]{1,2}/[0-9]{4})")
+parse_phone_number_regex = re.compile(r"\+?[0-9]*")
+
+
+def _find_and_parse(llm_answer, regex):
+    match = regex.findall(llm_answer)
+    return next(iter(match), None) if match else None
+
+
+def find_and_parse_date(llm_answer: str) -> str | None:
+    return _find_and_parse(llm_answer, parse_date_regex)
+
+
+def find_and_parse_phone_number(llm_answer: str):
+    return _find_and_parse(llm_answer, parse_phone_number_regex)
+

verification_prompt.txt

@@ -1,27 +0,0 @@
-Please tell if I answered the following questions:
-
-{questions}
-
-The answer should be a list of json objects that follow this schema and be legal json.
-{
-  "$schema": "https://json-schema.org/draft/2020-12/schema",
-  "$id": "https://www.enricorampazzo.tech/dam_helper.schema.json",
-  "title": "Answer",
-  "description": "an answer to a question",
-  "type": "object",
-  "properties": {
-    "question": {
-        "description": "the question being answered",
-        "type": "string"
-    },
-    "answer": {
-        "description": "the answer to the question",
-        "type": "string"
-
-    }
-  }
-}
-if the answer is a date format it as day/month/year.
-If the answer contains temporal references such as 'tomorrow', 'in two days' etc. consider that today it is the
-6th of september 2024.
-If the answer is not provided just the answer string should be null